Aesthetically beautiful futanari and traps.
TR5.8R (more poses (masturbation, sex) more retro, more realism).
TRW1.56CON Boosty (NEW VERSION - added chastity cage, condom on penis, greatly increased dataset, diversity and stability)
P.P.S. new model Anime Screencap / LORA / PONY DIFFUSION and NEW ERA 3.0 (MODEL) PONY DIFFUSION / SDXL on Boosty
For many arts, combinations of my lore were used, such as: Esthetic 90s style, Esthetic Urushihara Satoshi, Esthetic Lina Inverse (but new version from boosty)
I made a video on how to achieve the same quality or just repeat my art
Lora trained on almost 700 images with manual prompts settings. Few big cocks, mostly medium and small ones in the sample, because models like to make house-sized cocks, I don't like that.
changes in version TRW1.4CC:
added chastity cage
The dataset has been more than doubled
the quantity and quality of anime art has been increased, which should have a positive effect on the “anime” style
heavily reworked prompting
Eta noise seed delta 31337
Use model VividOrangeMix1.0_NSFW
or Abyssorangemix3aom3_aom3a3.safetensors model
or aom3a1
VERSION 5.0R much better with low weights. Use a weight around 0.2-.0.7, 0.8 and up is too "stylized" and too one-sided. Below 0.4 often makes a pussy, but you can try and add some basic key prompts: penis, futanari, otoko no ko, female focus, frenulum, small penis, large penis (carefully, but work), medium penis
flaccid, phimosis - helps with size reduction and do what they say
Negative promts: artist name (necessarily, most of the art with watermarks), (worst quality, low quality, extra digits:1.4), multiple penises, multiple hands, multiple testicles,
just use hires fix as much as your video card is capable of, so as not to give an error, with 8gb of video memory it is 512x768 with hires fix 2, upscaler Latent (nearest-exact) or just Latent (if there are problems with the image), the picture will be much more detailed.
CFG Scale - 5
I used arts mainly: exlic, tofuubear, personalami and many many others
The training took place at a resolution of 512x768, so it makes no sense to use 512x512
Combined with my other LORA
How to use LoRA's in auto1111:
Update webui (use
git pulllike here or redownload it)Copy the file to
stable-diffusion-webui/models/loraSelect your LoRA like in this video
Make sure to change the weight (by default it's
:1which is usually too high)
*Information taken from Lykon
page on Hugging Face https://huggingface.co/OneRing/TP3.7.safetensors
Please post your work with or without comments, it will help me improve. Thanks!
If you like my work, click on the heart above, I will be pleased :3
Description
If you want to support me, then you can subscribe to my patreon patreon.com/OneRing and get builds earlier than they come out here (versions on patreon will always be newer by one than here) This will help me a lot in updating the video card over time to make models even better and faster (the main thing is better)
First public release, works best with Grapefruit model (hentai model), but can also be used with AbyssOrangeMix2_hard and possibly anything-v4.5-pruned
FAQ
Comments (25)
can you share the lora config file?
@OneRing is this from
Aitrepreneur youtube?
@animesenpai69 almost, but there are some changes, just the location of the folders did not change
@OneRing how many image and steps do u use for training?
@animesenpai69 the steps are written, but in fact, by default, the number of steps grows with the number of photos, in my case, more than 100 photos with manual prompts (this is very long, but increases the quality, variety and accuracy) it turned out 10500 steps
Still experimenting with this, only complaint so far is that it seems to push its own styling onto the image, but if I decrease the strength too much then it just doesn't work well. Not sure how to remedy that.
Including masturbation with this would be great if it's possible.
At the moment, masturbation is not a priority and is difficult to implement, because models always have problems with their hands, but I will try to do something. At the moment, it is important to further improve the quality of the penises, increase their "frequency of occurrence" and add more angles.
@OneRing more angles would be useful, the underside rarely seems to gen properly.
The glans frequently doesn't render well if it's supposed to be exposed.
**I noticed while grabbing a few images to include with this that both of these issues seem to vary between models its used with and the strength, I'd need to mess with it more to give better feedback on that.
https://files.catbox.moe/hmsi1z.png
https://files.catbox.moe/kkyql8.png
https://files.catbox.moe/cs2536.png
https://files.catbox.moe/x9rlud.png
This looks great! Is there a chance for a ckpt version?
I’m afraid not, I have a rtx 3050 and no matter how much I try, I don’t have enough video memory, and if all the settings are cut to zero, the quality will drop and the processing time will be prohibitively high
@OneRing Ouh, that's a shame. Anyway thanks for creating this! I'm sure people who can run this are greatfull. :)
Hello, i'm new with stable diffusion, so I need some help.
how i can change the weight in stable diffusion web GUI, because is only generating pussy
After saving the LORA into the Lora folder, type in the prompt <lora:LORANAME:NUMBER>, for me its <lora:estheticFutanariTrap_tr25:0.4>
The number is taken from the description, on the AUTOMATIC1111 Gui there is also a button with this emoji "🎴" that lets you choose LORAs, Embeddings, Hypernetworks and Checkpoints
@FeverEncore Thank you! You helped so much
@yandesu Is that fixed for ya ? Still got many pussy
@Bayloz use more weight, important prompts or mix LORA files, the new version is ready, but I don't like it, many shortcomings
Hello OP, would U like to share your training data (fold of images)?
I didn’t quite understand the question, if you are talking about the settings, I already threw them, here is a slightly updated version - https://drive.google.com/file/d/1DJuyL9xK4_l90qoUVXrsUghvxj_PZqEg/view?usp=sharing Namely, I didn’t have a saved point, now I’m training again with already saved data for the future
@OneRing I think he wants a folder with all the pictures you used to train with.
@fongletto Thx & yes!
@OneRing I mean, may I get your training images? Thanks!
@losquit the selection of art took me more than a day, with the new version, which I have not posted yet - a few days, in the description there are authors from whom I took most of the works, you can select the ones you need yourself, you can add art axsens, aroma sensei, georgy stacker, I'm sorry, I would like to keep the exact sample a secret
Oh my god !! Thank you for this gift !!
A new version is being prepared (￣▽￣), 104 arts have been increased to 221, more poses have been added, back poses and their variety have been improved, old prompts have been fixed and new ones have been improved, in general, the variety should grow a lot and the quality in general. I also added some art of masturbation and penises under clothes, but it seems to me that this will not greatly affect the results so far, because these are difficult-to-implement aspects and we need more art and improve prompts, as well as lower the learning rate, which I cannot allow due to with my rtx3050, training already lasts half a day)
P.S. I have to re-learn due to a lot of changes+I forgot to save the model
(｡╯︵╰｡)